Report
Teaser, summary, work performed and final results
Periodic Reporting for period 2 - INQMINDS (The Evolutionary and Developmental Origins of Inquiring Minds: Studies of Causal Reasoning; Curiosity and Executive Control)
Teaser
Human technology is vastly superior to that of other apes: human tools from telescopes to the Large Hadron Collider exploit causal relationships but also explore them. What change over the brief course of hominid evolution made for such a big difference? One theory implicates...
Summary
Human technology is vastly superior to that of other apes: human tools from telescopes to the Large Hadron Collider exploit causal relationships but also explore them. What change over the brief course of hominid evolution made for such a big difference? One theory implicates a single cognitive distinction: the ability to reinterpret the world in abstract terms. In contrast, in several studies, chimpanzees seemed to rely on surface appearances, as if for the apes there was no more to the situation than met the eye. The theory is seductively simple, but this project argues that it is too early to rule out deep evolutionary roots for inquiring minds. lines of study: causal cognition and executive function. To disentangle causal reasoning from simpler mechanisms three approaches are planned. The first will look at intervention: will subjects use exploration to seek hidden causes? The second will focus on natural mechanics: do subjects reason about physical properties or rely on perceptual correlates? The third examines causal learning from observation when the task’s mechanics are opaque. The second line of study explores executive functions. An inquiring primate mind could flourish if more information could be held in mind and manipulated. But we know little about how executive functions compare across primates. A new test battery will explore levels of working memory, inhibition, and attention shifting. Executive functions undergo radical development in human childhood, and individual differences correlate with performance on tests of physical and social reasoning. The intersection between these skills may therefore be particularly revealing. This project aims to use cross-sectional testing of causal cognition and executive control in the same individuals to explore how these cognitive skills interact to produce inquiring minds. Our work will shed light on the fundamental building blocks of cognition and how these might change over evolution, with implications for evolutionary biology and artificial intelligence.
Work performed
We have made significant progress in the first part of the grant in line with what we had planned in the first 30 months – to design novel and rigorous tests of causal cognition and executive function that can be implemented across non-human primates and children and look for candidate tests that can be utilised in a large-scale cross-sectional test battery (planned for the second period). These are detailed below.
2 Principle Objectives
Work package 1 - Causal Cognition
We have been breaking new ground in the study of causal cognition along 3 main strands: Curiosity and Intervention, Natural mechanics and Computational modelling of observational learning. We have found interesting developmental trends in our work with children and large individual differences which is very promising for work package 3 – the cross-sectional study. We have found largely negative results with non-human primates so far, which is very interesting given that the methodology has been designed to be extremely rigorous and to rule out simpler explanation for success more robustly. Negative results are much harder to interpret than positive ones, and so we are focusing on triangulating on these results with new methods to validate the species differences. We may need to rethink what tasks we include in the nonhuman primate test-battery in the next period of the grant.
Curiosity and Intervention
We developed a new paradigm with pre-schoolers which showed that in the absence of verbal framing, children could use free exploration to learn about object properties and solve a task. We are now testing non-human primates with the same methodology.
Natural mechanics
We developed new paradigms to compare learning based on the same cues in two different scenarios: one causal, in which the cues have mechanical relevance to the outcome, and one arbitrary, in which although the same cue is 100% predictive, it has no mechanical relevance to the outcome. These respond to previous criticisms because the causal context is only differentiated from the arbitrary one by the order of two predictive cues: seeing a reward drop behind an occluder then hearing a sound that predicts where it will emerge gives the illusion of causality, while hearing the predictive sound before the reward is dropped does not. We found that while 4-6 year-olds children performed well in the causal task, 3 year olds, chimpanzees and capuchin monkeys did not. We are now extending this approach to another paradigm to validate this finding.
We also implemented a task in which the same feature has to be treated differently depending on the causal role it plays: steering food away from a block when it serves as a barrier, but towards it when it serves as a supporting surface. There had been no systematic comparisons of apes, monkeys and humans on these kinds of tasks. We found no evidence that capuchins could solve this task, while chimpanzees and children performed well.
Computational modelling of observational learning
In collaboration with Dr. Daphna Buchsbaum in Toronto as described in the description of the action, we have been developing probabilistic computational models using Bayesian inference to explore children and non-human primates’ social and causal inferences from demonstrations, in order to gain insight into the potential mechanisms underlying differences in performance. We have found that older children integrate social and physical information in deciding what to copy, but that capuchin monkeys and younger children seem only to take physical information into account. We are verifying this pattern of findings with a second methodology, and preparing our results for publication.
We have also been exploring how capuchin monkeys and children learn from sparse sampling data. While we found evidence that children form overhypotheses about the nature of the population, we found no such evidence for monkeys. We are exploring this species difference further with a sec
Final results
In all areas of the action we are developing novel paradigms and breaking new ground in the area of comparative psychology. We have written a theoretical paper which contributed to a special issue in examining individual differences, which explains the rationale for our multi-trait multi-method test battery and the importance of this approach in understanding both individual and species differences in cognition. This paper is under revision at Philosophical Transactions of the Royal Society B.
The next step is to combine our new test into a test battery, to explore how executive functions relate across primate species, and the possible consequences of individual differences in executive function for other cognitive skills, such as causal cognition and observational learning.
Our work sits at the boundary of comparative, developmental and cognitive psychology, computational science and evolutionary biology. The PI presented at a cross-disciplinary workshop on Human-like computing and is now part of a network of computational and cognitive scientists trying to understand how to implement human-like computing in artificial intelligence, EPSRC Network, on Human-Like Computing, which will run for five years starting in April 2018. We also participated in the above mentioned cross-disciplinary workshop on individual differences which brought together behavioural ecologists, comparative psychologists and physiologists. Our work will shed light on the fundamental building blocks of cognition and how these might change over evolution, with implications for evolutionary biology and artificial intelligence.